EA Survey: Sexual Harassment Questions—Feedback Requested

I in­ter­ested Tee Bar­nett and Peter Hur­ford in adding sex­ual vi­o­lence ques­tions to the sur­vey. There­fore sex­ual vi­o­lence defi­ni­tions need to be cre­ated. There are var­i­ous challenges in­volved. Below I have listed the challenges and pro­posed a set of ques­tions that I think will help work around them. (This does not mean Tee and Peter will in­clude my par­tic­u­lar ques­tions. I am mak­ing a pro­posal and out­lin­ing spe­cific challenges that I am aware of.)

    1. Sur­vivor­ship bias. The sur­vey will only be taken by peo­ple who are still around, not by peo­ple who have left their EA job or been driven out of the com­mu­nity by sex­ual vi­o­lence. Sex­ual vi­o­lence does in­crease turnover. Given what hap­pens to fe­male lawyers in the le­gal pro­fes­sion, which has a similar gen­der ra­tio, we should be very con­cerned about this. Ac­cord­ing to The Amer­i­can Bar As­so­ci­a­tion, be­tween half and two thirds of fe­male lawyers ex­pe­rienced or ob­served sex­ual ha­rass­ment by male su­pe­ri­ors, col­leagues, or clients dur­ing the two years prior to the sur­vey. [1] There are some other refer­ences on turnover as well. [2] [3] [4] [5] [6]

      If you want a solid ex­am­ple, I am leav­ing you. That’s right. I plan to exit the effec­tive al­tru­ism net­work, the LessWrong di­as­pora, and all of my male-dom­i­nated in­ter­ests in­clud­ing pro­gram­ming. I am sick and tired of sex­ual ha­rass­ment, sick and tired of peo­ple be­hav­ing as if they don’t un­der­stand or care about my phys­i­cal safety con­cerns, and I’m sick and tired of en­coun­ter­ing dis­mis­sive and in­sen­si­tive at­ti­tudes about some­thing that is caus­ing me suffer­ing and harm. I am leav­ing.

      A lot of other women have left you. They’ve com­plained in the women’s group. They’ve warned me in pri­vate mes­sages. They’ve com­mented and urged me to join them in this that and the other to­tally un­re­lated com­mu­nity. They were do­ing these kinds of things *be­fore* I pub­lished my ar­ti­cle “Ex­plo­ra­tion of Sex­ual Violence Re­duc­tion for Effec­tive Altru­ism Po­ten­tial”. Some peo­ple seem to want to avoid talk­ing about this to pre­vent women from be­ing scared off. Women are be­ing tar­geted by sex offen­ders. They are go­ing to no­tice that. Sex offen­ders are scar­ing peo­ple off whether we talk about it or not.

      If we want to be rigor­ous enough to have mean­ingful sur­vey re­sults, we *need* to com­pen­sate for sur­vivor­ship bias. I have no doubts about that.

      Lists of EA Global at­ten­dees from differ­ent years can be com­pared to see how many peo­ple re­turned and how many peo­ple did not. CEA won’t re­lease their data. There is a quick and dirty way to ac­com­plish a rough es­ti­mate with­out the data. CEA told me they’d run a script that:

      1. Com­pares the full names of at­ten­dees of EA Global from two differ­ent years to see which peo­ple re­turned to the con­fer­ence and which peo­ple did not. (Keep­ing the names a se­cret.)

      2. Counts the num­ber of “male” and “fe­male” names (and am­bigu­ous names) on each side. (Based on a database of baby names which in­cludes the gen­der they’re usu­ally as­signed to.)

      3. Out­puts a table con­tain­ing the to­tal num­ber of names in each cat­e­gory that did and did not at­tend both events.

An r or python script can be run in an in­stal­la­tion of r statis­tics on one of the CEA com­put­ers. *Only* the table con­tain­ing the com­pletely anonymized nu­mer­i­cal to­tals needs to be shared. So yes, this is quick and dirty and would only out­put a rough es­ti­mate. The only way to get re­ally ac­cu­rate data is for CEA to have an in­ter­nal per­son write the script.

    1. Defi­ni­tional quag­mires. Sadly, the defi­ni­tions of sex­ual vi­o­lence vary a lot from one source to the next, and this even ap­plies to the sources that are sup­posed to be cred­ible like le­gal defi­ni­tions, psy­chol­ogy defi­ni­tions, and re­search defi­ni­tions. Worse, all of the defi­ni­tions I have seen so far are com­pletely in­sane in at least one re­spect. For in­stance, the CDC defi­ni­tion clas­sifies male rape via pe­nile en­velop­ment as a differ­ent kind of sex offence, as if it is not rape. It is *not* in­cluded in the num­ber of rapes! For some rea­son, a lot of defi­ni­tions of stalk­ing re­quire that the per­son stalks you more than once. What if the stalker fol­lows you around and suc­cess­fully at­tacks you on the first try? I guess some­one de­cided that when stalk­ers are es­pe­cially effec­tive at what they do, they aren’t re­ally stalk­ing af­ter all? In­sane. A lot of the defi­ni­tions I’ve seen con­tain some­thing nutty like this.

      1. No stan­dard­ized defi­ni­tions.

      2. Lots of defi­ni­tions in­clude at least one re­ally in­sane loop­hole.

      3. All the terms are am­bigu­ous be­cause a lot of peo­ple re­fuse to re­port sex offences if they have to be spe­cific. This is rem­i­nis­cent of what hap­pened to user agent val­ues in the browser wars though it hap­pens for a differ­ent rea­son. (They started hav­ing browsers lie to web­sites about which browser they are as a chaotic way to man­age com­pe­ti­tions over fea­tures and com­pat­i­bil­ity, so the iden­ti­fiers be­came pretty mud­dled.)

      4. Hu­mans re­spond to in­cen­tives with mo­ti­vated rea­son­ing, cog­ni­tive bi­ases and other forms of ir­ra­tional­ity. This means that the defi­ni­tions we use must be re­sis­tant to peo­ple com­ing along and to tear them down sim­ply be­cause a state of de­nial is a lot more pleas­ant.

      5. Sex­ual vi­o­lence defi­ni­tions do not have a util­i­tar­ian ba­sis like how much suffer­ing they cause. This is prob­a­bly be­cause there can be many differ­ences be­tween an event that is up­set­ting vs. an event which does not. If a dog sniffs your crotch, you will prob­a­bly be an­noyed but not trau­ma­tized. If an adult hu­man did that, it would be very up­set­ting. Why? The dog is smaller. The dog prob­a­bly isn’t ex­press­ing in­ter­est in you. The dog has no poli­ti­cal power what­so­ever. The dog may be ex­press­ing friendly emo­tions rather than sadism, dom­i­nance, etc. Prob­a­bly other rea­sons. There are many con­tex­tual things like this which re­sult in a huge amount of con­fu­sion and also make it ex­tremely hard to define sex­ual vi­o­lence in a util­i­tar­ian way. A lot of peo­ple in our net­work are in­ter­ested in whether ev­ery type of be­hav­ior that qual­ifies as a sex offence causes harm.

      6. The main benefit of uti­liz­ing a com­mon set of sex­ual vi­o­lence defi­ni­tions is that we’ll eas­ily be able to com­pare our rate of sex offences to ex­ter­nal rates of sex offences to know whether it is ele­vated or not. The main down­side of uti­liz­ing a com­monly used set of sex­ual vi­o­lence defi­ni­tions is that most of them in­clude some­thing in­sane. Un­less we can find a cred­ible source of sex­ual vi­o­lence defi­ni­tions which has no in­sane loop­holes, we will need to choose be­tween com­pa­ra­bil­ity and san­ity.

    2. In­suffi­cient un­der­stand­ing of sex­ual re­sponses and prefer­ences: Some peo­ple never ex­pe­rience sex­ual trauma, or are not as likely to ex­pe­rience it due to some kind of trait. Per­haps they did not re­ceive the genes for trauma at all, or the genes for sex­ual trauma speci­fi­cally, or they ex­pe­rience sex­ual trau­mas un­der a to­tally differ­ent set of con­di­tions from oth­ers.

      From what I’ve been see­ing, I am re­ally start­ing to sus­pect that there are mul­ti­ple “ori­en­ta­tions” around con­sent needs and what situ­a­tions are ex­pe­rienced as sex­ual vi­o­lence. Some peo­ple ve­he­mently deny that sex offences cause harm or act like harm is rare when re­search shows it is not. I think what’s hap­pen­ing is that peo­ple with non-stan­dard sex­ual re­sponses and sex­ual prefer­ences tend to con­struct bub­bles based on these traits. They may not re­al­ize they are liv­ing in a bub­ble and that the rest of us re­ally do ex­pe­rience sex­ual trauma.

      There isn’t a defi­ni­tion that such peo­ple will find cred­ible be­cause the crux is not about ver­biage for them. What they would find cred­ible is an ex­pla­na­tion as to why some peo­ple claim X is harm­ful while oth­ers claim X was en­joy­able or did not cause harm.

      My best an­swer for them is that most peo­ple prob­a­bly ex­pe­rience sex­ual trauma, while those who don’t ex­pe­rience it (or who have differ­ent con­di­tions for it) prob­a­bly at least some­times end up in so­cial bub­bles with oth­ers who have differ­ent sex­ual re­sponses and be­hav­iors.

      I don’t know all the de­tails, and I may only see the tip of the ice­berg. Right now, I’m pretty sure that some peo­ple need both ex­plicit ver­bal con­sent and mon­i­tor­ing for en­thu­si­asm in or­der to func­tion sex­u­ally and avoid trauma. Others re­quire body lan­guage (Or per­haps other cues? I am not one of these peo­ple.) to func­tion sex­u­ally, and have liter­ally re­ported that us­ing ex­plicit ver­bal con­sent ac­tu­ally causes them dys­func­tion. One per­son re­ported that this is *in spite of* the risk of trauma, *not* be­cause the risk of trauma is ab­sent.

      To make it more com­pli­cated, I think there are kinky peo­ple in both the ex­plicit con­sent and body lan­guage camps. That is to say I think some kinky peo­ple sex­u­ally dys­func­tion if they can’t use body lan­guage and must use ex­plicit ver­bal con­sent.

      Such peo­ple are not go­ing to cease their sex­ual be­hav­iors be­cause oth­ers func­tion a differ­ent way. They wouldn’t func­tion at all then. At this time, I think it is safest for ev­ery­one to en­courage peo­ple to form differ­ent bub­bles that all mem­bers are com­fortable with. Vilify­ing com­fortable bub­ble cre­ation doesn’t make much sense. I think the peo­ple in each bub­ble just need to be aware that there are peo­ple in var­i­ous other bub­bles who gen­uinely have differ­ent re­sponses and prefer­ences.

      I think right now there are a lot of mi­s­un­der­stand­ings be­cause the peo­ple in the differ­ent bub­bles don’t even be­lieve that the other bub­bles re­ally ex­ist for sure, let alone do peo­ple from each bub­ble know how the other peo­ple work. We’ve all been view­ing each other’s bub­bles as if they are non-ex­is­tent, imag­i­nary or dys­func­tional.

      I think if peo­ple shared openly the con­sent philos­o­phy that they pre­fer oth­ers to use on them, this would help im­mensely with pre­vent­ing con­fu­sion when peo­ple in mul­ti­ple bub­bles mix. I think this has the po­ten­tial to re­duce sex­ual vi­o­lence be­cause it will re­duce con­fu­sion be­tween those who need ex­plicit ver­bal con­sent and those who need to use body lan­guage. There may be a lot of other differ­ences I am un­aware of, so I will leave off with a hope that peo­ple will com­mu­ni­cate more clearly in the fu­ture to pre­vent mi­s­un­der­stand­ings.


      Bub­bles are *not* a com­plete solu­tion. This is be­cause some peo­ple liter­ally have prefer­ences like “do­ing stuff you’re not sup­posed to do” /​ “do­ing un­wanted things”. Yes, I do mean that there are peo­ple who do this *with­out* hop­ing to delight the other per­son. (Some sadists and dom­i­nants pre­fer to delight masochists and sub­mis­sives while oth­ers just go for any­body with­out con­sid­er­a­tion, and some ac­tu­ally in­tend to cause peo­ple harm.)

      Others have prob­lems along the lines of a strong mind pro­jec­tion fal­lacy which re­sults in them be­liev­ing oth­ers re­cip­ro­cate their in­ter­est when this is not true. This bias might be bad enough to cause a sex offence. Ad­di­tion­ally, it seems that many peo­ple ex­pe­rience so much mo­ti­vated rea­son­ing that they rape peo­ple even though they ex­pe­rience em­pa­thy and should the­o­ret­i­cally be able to tell that they’re do­ing some­thing harm­ful. I guess, for some, the mo­ti­vated rea­son­ing and bias that can hap­pen dur­ing mat­ing ex­pe­riences is deeply con­fus­ing.

      Then there are peo­ple who take more risk with sex or kink than they in­tend to and some­one ends up un­nec­es­sar­ily harmed through a lack of ed­u­ca­tion. So, un­for­tu­nately, there are a bunch of things that can re­sult in peo­ple get­ting hurt even if ev­ery­one aims to com­mu­ni­cate re­ally well and peo­ple stick to dat­ing/​hookups/​play with oth­ers who share a com­pat­i­ble con­sent philos­o­phy.

      If we use a main­stream source of sex­ual vi­o­lence defi­ni­tions, it will stig­ma­tize peo­ple who need to use body lan­guage rather than ex­plicit ver­bal con­sent. They prob­a­bly aren’t go­ing to change their be­hav­ior, since that would re­sult in sex­ual dys­func­tion. Also, I’m con­cerned that there prob­a­bly is not enough re­search on peo­ple with this prefer­ence for us to have any un­der­stand­ing of them. Without thor­ough re­search, stigma is un­jus­tified.

      If we don’t use defi­ni­tions that re­quire ex­plicit ver­bal con­sent, this may open the floodgates for a lot of peo­ple who are con­fused by mo­ti­vated rea­son­ing and bias. It is prob­a­bly safer if peo­ple prone to mo­ti­vated rea­son­ing and bias do *not* try to use body lan­guage to figure out what oth­ers want. They need to work ex­tra hard to make dis­tinc­tions and I sus­pect it helps them to be told clearly and ex­plic­itly what is wanted.

    3. Cog­ni­tive bi­ases, mo­ti­vated rea­son­ing, and in­cen­tives.

      1. Op­ti­mism bias:
        The ten­dency to ex­pect things to be more pleas­ant than they are.

      2. Nor­malcy bias:
        The ten­dency to fail to plan in ad­vance for dis­asters and to re­spond in­ap­pro­pri­ately when one is un­der­way.

      3. Ostrich effect bias:
        Ig­nor­ing an ob­vi­ous nega­tive situ­a­tion.

      4. Affect heuris­tic:
        Be­ing more likely to be­lieve in­for­ma­tion that is pleas­ant than in­for­ma­tion that is un­pleas­ant.

      5. Sta­tus quo bias:
        Be­ing more likely to sup­port the sta­tus quo.

      6. Sur­vivor­ship bias:
        Fo­cus­ing on who or what is left /​ failing to take into ac­count who or what is gone.

      7. Sour grapes bias:
        Tel­ling your­self that what you’ve lost wasn’t de­sir­able any­way.

Pro­posed defi­ni­tions:

The things I care about the most are things like whether some­one is trau­ma­tized, whether they be­come suici­dal and whether they leave or in­tend to leave the move­ment or their job.

What would be *re­ally* in­ter­est­ing is to ask *both* “did things A, B, or C hap­pen to you” *and* “did it cause you to ex­pe­rience X, Y, Z harm”?

This would be long be­cause there’d be a sub­set of ques­tions *about* each of the ques­tions in the main set. Also, since mul­ti­ple offences can hap­pen to one per­son, you’d need to be able to an­swer each ques­tion and sub­set ques­tion more than one time. This sounds com­pli­cated and I’m not sure it will hap­pen since sur­veys have length con­straints.

This would be awe­some be­cause we would have an op­por­tu­nity to un­der­stand this so much bet­ter by tak­ing a peek at which be­hav­iors cause the most harm.

If it needs to be con­cise, I’m imag­in­ing some­thing like this:

      1. What con­sent philos­o­phy do you pre­fer oth­ers use on you? (If you use a par­tic­u­lar vari­a­tion on a con­sent philos­o­phy, please spec­ify.)

      2. Where is your con­sent philos­o­phy doc­u­mented? (Please link to the spe­cific vari­a­tion if pos­si­ble.)

      3. How many times did any EA vi­o­late your con­sent philos­o­phy with speech that ap­peared to be mo­ti­vated by sex­ual or ro­man­tic de­sire?

      4. How many times did any EA vi­o­late your con­sent philos­o­phy with ex­hi­bi­tion­ism that ap­peared to be mo­ti­vated by sex­ual or ro­man­tic de­sire?

      5. How many times did any EA vi­o­late your con­sent philos­o­phy with voyeurism that ap­peared to be mo­ti­vated by sex­ual or ro­man­tic de­sire?

      6. How many times did any EA vi­o­late your con­sent philos­o­phy with touch­ing that ap­peared to be mo­ti­vated by sex­ual or ro­man­tic de­sire?

      7. How many times did any EA vi­o­late your con­sent philos­o­phy with fol­low­ing, in­trud­ing or spy­ing (non voyeurism) that ap­peared to be mo­ti­vated by sex­ual or ro­man­tic de­sire?

      8. How many of the afore­men­tioned in­stances of vi­o­la­tions had a sig­nifi­cant nega­tive im­pact on your psy­cholog­i­cal health?

      9. How many of the afore­men­tioned in­stances of vi­o­la­tions re­sulted in suici­dal thoughts?

      10. How many of the afore­men­tioned in­stances of vi­o­la­tions re­sulted in a de­sire to leave an EA job?

      11. How many of the afore­men­tioned in­stances of vi­o­la­tions re­sulted in a de­sire to leave the EA move­ment?

      12. If ap­pli­ca­ble, please list all the types of EA spaces where the afore­men­tioned in­stances of sex­ual ha­rass­ment or sex­ual vi­o­lence hap­pened. (ex­am­ples: EA share house, EA or­ga­ni­za­tion, EA con­fer­ence, EA work party, an un­re­lated cafe, a park, etc.).

      13. What do you think might have pre­vented the vi­o­la­tion(s) of your con­sent philos­o­phy?

This wouldn’t give us the sort of re­sult that’s di­rectly com­pa­rable to ex­ter­nal sex­ual vi­o­lence statis­tics. How­ever, it would give us the sort of re­sult we care about: whether peo­ple’s preferred con­sent philoso­phies are be­ing vi­o­lated, and whether the vi­o­la­tions are caus­ing harm. Most im­por­tantly, we’d ac­tu­ally have some data about the im­pact on spe­cific ar­eas in­clud­ing psy­cholog­i­cal health, suici­dal thoughts, turnover in­ten­tions, and in­ten­tions to leave the com­mu­nity.

The re­sults would be harder to com­pare from one year to the next be­cause ty­ing sex­ual vi­o­lence defi­ni­tions to peo­ple’s in­di­vi­d­ual ideas about con­sent adds more com­plex­ity than bru­tal­iz­ing all the sub­tleties to cre­ate stan­dard­ized ver­biage. It *is* pos­si­ble to make sur­vey re­sults com­pa­rable *if* peo­ple put in the effort to thor­oughly doc­u­ment all the differ­ent sets of prefer­ences they have, name each con­sent philos­o­phy or vari­a­tion, and pub­lish these. I think en­courag­ing peo­ple to do this is a good thing be­cause it en­courages us to deepen our un­der­stand­ing and com­mu­ni­cate more clearly.

Ty­ing the defi­ni­tion to peo­ple’s in­di­vi­d­ual prefer­ences just seems a lot more sane than try­ing to come up with sex­ual vi­o­lence defi­ni­tions that work for ev­ery­body at a time when so much more re­search needs to be done to un­der­stand all the vari­a­tion in re­sponses and prefer­ences, not to men­tion solv­ing what­ever it is that has been re­sult­ing in in­sane loop­holes.

Nonethe­less, I think some­thing in this gen­eral di­rec­tion would be much bet­ter at track­ing the things we care about the most: whether prefer­ences are be­ing re­spected, and whether harm is be­ing done.

I’m not an ex­pert on de­sign­ing sur­veys. There will be peo­ple on the sur­vey team look­ing this over and it may need to be changed. This is step one in the pro­cess: post a draft to get feed­back on!

Please provide your feed­back.


    1. La­band, David N., and Bernard F. Lentz. “The effects of sex­ual ha­rass­ment on job satis­fac­tion, earn­ings, and turnover among fe­male lawyers.” ILR Re­view 51.4 (1998): 594-607.


    2. Sims, Carra S., Fritz Dras­gow, and Louise F. Fitzger­ald. “The effects of sex­ual ha­rass­ment on turnover in the mil­i­tary: time-de­pen­dent mod­el­ing.” Jour­nal of Ap­plied Psy­chol­ogy 90.6 (2005): 1141.


    3. Merkin, Re­becca S. “The im­pact of sex­ual ha­rass­ment on turnover in­ten­tions, ab­sen­teeism, and job satis­fac­tion: Find­ings from Ar­gentina, Brazil and Chile.” Jour­nal of In­ter­na­tional Women’s Stud­ies 10.2 (2008): 73.


    4. Merkin, Re­becca S., and Muham­mad Ka­mal Shah. “The im­pact of sex­ual ha­rass­ment on job satis­fac­tion, turnover in­ten­tions, and ab­sen­teeism: find­ings from Pak­istan com­pared to the United States.” SpringerPlus 3.1 (2014): 215.


    5. Sal­man, Ma­heen, Fa­had Ab­dul­lah, and Afia Saleem. “Sex­ual Harass­ment at Work­place and its Im­pact on Em­ployee Turnover In­ten­tions.” Busi­ness & Eco­nomic Re­view 8.1 (2016): 87-102.


    6. San­dada, Maxwell. “The in­fluences of sex­ual ha­rass­ment on health, psy­cholog­i­cal con­di­tion, work with­drawal and turnover in­ten­tion in South Africa.” Jour­nal of Busi­ness 1.2 (2013): 84-72.